Optimality
Optimality
Optimality
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Semiparametric transformation models 141<br />
the case of generalized proportional hazards intensity models. Using results of section<br />
6, we can show that the sample analogue of the function ˙ Γθ satisfies Conditions<br />
2.3(iv) and 2.3(v).<br />
Example 2.3. The logarithmic derivatives of ℓ(x, θ, Z) = log α(x, θ, Z) may be<br />
difficult to compute in some models, so we can try to replace them by different<br />
functions. In particular, suppose that h(x, θ, Z) is a differentiable function with<br />
respect to both arguments and the derivatives satisfy a similar Lipschitz continuity<br />
assumption as in condition 2.1. Consider the score process (2.6) with function<br />
ϕθ = 0 and weights b1i(x, t, θ) = h(x, θ, Zi)−[Sh/S](x, θ, t) where Sh(x, θ, t) =<br />
� n<br />
i=1 Yi(u)[hα](x, θ, Zi), and ϕnθ≡ 0. For p = 0 and p = 2, define matrices Σ h pϕ by<br />
replacing the functions vϕ and ρϕ appearing in matrices Σ0ϕ and Σ2ϕ with<br />
v h ϕ(t, θ0) = var[h(Γθ0(X), θ0, Z)|X = t, δ = 1],<br />
ρ h ϕ(t, θ0) = cov[h(Γθ0(Xi), θ0, Zi), ℓ ′ (Γθ0(Xi), θ0, Zi)|X = t, δ = 1].<br />
The matrix Σ1ϕ(θ0, τ) is changed to Σ h 1ϕ(θ0, τ) = � ¯ρ h ϕ(t, θ0)EN(du), where the<br />
integrand is equal to<br />
cov[h(Γθ0(X), θ0, Z), ˙ ℓ(Γθ0(X), θ0, Z) + ℓ ′ (Γθ0(X), θ0, Z) ˙ Γθ0(X)|X = t, δ = 1].<br />
The statement of Proposition 2.2 remains valid with matrices Σpϕ replaced by<br />
Σ h pϕ, p = 1, 2, provided in Condition 2.3 we assume that the matrix Σ h 0ϕ is positive<br />
definite and the matrix Σ h 1ϕ is non-singular. The resulting estimates have a structure<br />
analogous to that of the M-estimates considered in the case of uncensored data by<br />
Bickel et al. [6] and Cuzick [13]. Alternatively, instead of functions ˙ ℓi(x, θ, z) and<br />
ℓ ′ (x, θ, z), the weight functions b1i and b2i can use logarithmic derivatives of a<br />
different distribution, with the same parameter θ. The asymptotic variance is of<br />
similar form as above. In both cases, the derivations are similar to Section 7, so we<br />
do not consider analysis of these score processes in any detail.<br />
Example 2.4. Our final example shows that we can choose the ϕθ function so that<br />
the asymptotic variance of the estimate ˆ θ is equal to the inverse of the asymptotic<br />
variance of the normalized score process, √ nUn(θ0). Remark 2.1 implies that if<br />
ρ−Γ ˙ (u, θ0)≡0 but v(u, θ0)�≡ 0 a.e. EN, then for ϕθ =− ˙ Γθ the matrices Σq,ϕ, q =<br />
1,2 are equal. This also holds for v(u, θ0)≡0. We shall consider now the case<br />
v(u, θ0)�≡ 0 and ρ−Γ ˙ (u, θ0)�≡ 0 a.e. EN, and without loss of generality, we shall<br />
assume that the parameter θ is one dimensional.<br />
We shall show below that the equation<br />
(2.7)<br />
ϕθ(t) +<br />
� τ<br />
0<br />
=− ˙ Γθ(t) +<br />
Kθ(t, u)v(u, θ)ϕθ(u)EN(du)<br />
� τ<br />
0<br />
Kθ(t, u)ρ(u, θ)EN(du)<br />
has a unique solution ϕθ square integrable with respect to the measure (2.4). For θ =<br />
θ0, the corresponding matrices Σ1,ϕ(θ0, τ) and Σ2,ϕ(θ0, τ) are finite. Substitution<br />
of the conditional correlation function ρϕ(t, θ0) = ρ(t, θ0)−ϕθ0(t)v(t, θ0) into the<br />
matrix Σ2,ϕ(θ0, τ) shows that they are also equal. (In the multiparameter case, the<br />
equation (2.7) is solved for each component of the θ).<br />
Equation (2.7) simplifies if we replace the function ϕθ by ψθ = ϕθ + ˙ Γθ. We get<br />
� τ<br />
(2.8) ψθ(t)−λ<br />
0<br />
Kθ(t, u)ψθ(u)Bθ(du) = ηθ(t),